EAA2020: Abstract
Abstract is part of session #464:
Title & Content
Title:
More than the eye can see - Machine Learning and rock art
Content:
In recent years, over 400 laser scans of rock art panels in Sweden have been produced of rock art dating to the Nordic Bronze Age (1800-550 BC). Among the many thousands of motifs are boats, animals, wagons, humans, etc. In 2019, the Swedish Rock Art Research Archives and its partner institutions started a project that attempts to use the 3D data and automatically classify motifs on the panels. The classification of this material can eventually be used for quantitative approaches to the study of rock art. To this end, an initial convolutional neural network (Faster RCNN) has been trained successfully. This could be used in new projects for statistical analysis and potentially at some point identify individual carvers.
This talk will present the method and preliminary results of the project. In the predictions, we encountered unexpected results and difficulties highlighting the ambiguity of Scandinavian rock art which includes flowing forms, abstractions, and hybrids. Based on these findings, we will discuss the relationship between human creativity and computer vision.
This talk will present the method and preliminary results of the project. In the predictions, we encountered unexpected results and difficulties highlighting the ambiguity of Scandinavian rock art which includes flowing forms, abstractions, and hybrids. Based on these findings, we will discuss the relationship between human creativity and computer vision.
Keywords:
Machine Learning, Rock Art, Computer Vision, Human Vision
Downloads:
authors
Main authors:
Christian Horn1
Co-author:
Johan Ling2
Oscar Ivarsson3
Oscar Ivarsson3
Affiliations:
1 Gothenburg University
2 University of Gothenburg
3 Chalmers University of Technology
2 University of Gothenburg
3 Chalmers University of Technology